Post on 08-Apr-2018
transcript
Visit our website: www.earth2observe.eu
This project has received funding from the European Union’s Seventh Programme for researchtechnological development and demonstration under grant agreement No 603608
Earth2Observe
Are global models skilful in forecasting floods, and their impacts in data scarce areas?
Micha Werner (1,2), Gaby Gründermann (1), Ted Veldkamp (3)
(1) IHE Delft, Department of Water Science and Engineering, The Netherlands
(2) Deltares, The Netherlands
(3) Free University Amsterdam, Amsterdam, the Netherlands
HEPEX 2018University of Melbourne
Motivation
Motivation
• Global models have potential for assessment and prediction of flood
hazard in areas with insufficient data
– Asymmetric availability of data (transboundary basins)
– Period of record of consistent hydrological data short
• But…
– How good are these models in predicting floods and their impacts?
– What about scale (basin scale, resolution of hydrological model)?
Lim
po
po
Riv
er a
t X
aiX
ai, M
oza
mb
iqu
e. P
ho
to: K
arel
Pri
nsl
oo
Sto
rm o
ver
Joh
ann
esb
urg
. P
ho
to: D
lala
Nje
Approach
• Case Study: Limpopo Basin in Southern Africa
– South Africa; Botswana; Zimbabwe; Mozambique
• Selection of global models from EartH2Observe Water Resources
Reanalysis (WRR) that include simulated discharge
– WRR1: Resolution 0.5 degrees; Daily; Forced by WFDEI Dataset; 1979-2012
– WRR2: Resolution 0.25 degrees; Daily; Forced by MSWEP Dataset; 1980-2014
• Comparison against 2 Benchmarks
A: Observed discharges at (reliable) discharge stations across basin
B: Chronology of impacting flood events from disaster databases
Approach
Benchmark A. Observed discharges
Limpopo (98240 km2)
Spookspruit (252 km2)
72 Stations
Performance of
simulated discharge
Flood Severity Level
ZA1
ZA2
BW
ZW
MZZA3
ZA4
Benchmark B. Reported impacting flood events
• EM-DAT – (CRED & Guha-Sapir, 2017)
• GAALFE – Dartmouth Flood Observatory (Brakenridge, 2017)
• NatCatSERVICE –Munich Re (Kron et al., 2012)
• Severity Level 0-5 based on NatCatSERVICE amended for no. of casualties / Basin
Level
Benchmark B. Reported impacting flood events
• EM-DAT – (CRED & Guha-Sapir, 2017)
• GAALFE – Dartmouth Flood Observatory (Brakenridge, 2017)
• NatCatSERVICE –Munich Re (Kron et al., 2012)
• Severity Level 0-5 based on NatCatSERVICE amended for no. of casualties
• Sub Basin/Country Level
Model performance
NSE PBIAS Correlation
WRR20.25 deg
WRR10.5 deg
Occurrence of Flood Events
Example for WaterGAP model at Spookspruit & Limpopo gauges
Flood events identified using model climatology (MM1 & MM2)
Flood events identified using observed climatology (MO1 & MO2)
Digit indicates model resolution; 1 - WRR1 (0.5 degrees); 2 – WRR2 (0.25 degrees)
Occurrence of Flood Events (against observed)
CSI; POD & FAR using Annual exceedance probability threshold of 0.164 (5 years return
period) for all gauging stations. WRR1 (upper panel) & WRR2 (lower panel).
CSI POD FAR
WRR20.25 deg
WRR10.5 deg
Simulated return periods of reported flood events
The relationship of the flood event severity for the reported flood events, and the corresponding
annual exceedance probabilities that were observed and modelled for (a) HTESSEL-CaMa, (b)
LISFLOOD, and (c) WaterGAP3.
Discussion & Conclusions
• Overall performance of global models in simulating hydrological behaviour
rather poor for smaller catchments
– WRR1 basic representation of hydrological behaviour > ~2500 km2
– WRR2 basic representation of hydrological behaviour > ~520 km2
• Skill of identifying observed flood events reasonable – but only when
using model climatology.
• Models also show some skill in identifying flood events that cause impacts
– important for their use in e.g. global forecasting systems
– Improves for improved resolution WRR2 models (with exceptions)
• Global models provide information consistently – also for transboundary
basins with asymmetric data availability
• Caveats: Inclusion of human influences in models and data; reliability of
gauged discharges, particularly at peaks